The Impact of Radiotherapy Dose on Local Control of Ewing's Sarcoma of Bone

Purpose. Improvements in the systemic management of Ewing's sarcoma of bone over the last 20 years have led to a dramatic improvement in survival. The corollary is that treatment of the primary disease requires re-evaluation, since a significant number of patients still suffer local relapse. Patients. The effect of radiation dose on local control was reviewed in a series of 96 patients treated between 1967 and 1986. Seventy-four had no metastases at presentation (M0), 22 had metastases (M1). The 5-year survival of all patients was 28%, and of M0 patients alone 37%. Although these figures are poor by today's standards, they are consistent with published studies whose patients were enrolled during the same calendar period. Although most deaths occurred by 5 years, survival continued to fall beyond 10 years, which has implications for follow-up in future studies. Results. The local control (LC) rate at 5 years was 56% for all patients and for M0 patients analyzed separately. There was no difference in either LC or survival between the first and second decades of the study. Primary site was a significant determinant of survival and local control, with better outcome for limb tumours compared to pelvic primaries. Chemotherapy also had a major effect on LC. Radiotherapy improved the probability of LC. Omission of radiotherapy, or a dose <40 Gy, was ineffective. In the dose range 40–66 Gy, there was no evidence of a dose–response relationship.


Introduction
Ew ing' s sarcom a is the second com m onest tum our of bone in children with an annual U K incidence of approxim ately 1.7 cases per m illion, representing 1± 2% of childhood tum ours, 1 and betw een 6± 15% of all primary bone tum ours. 2 It arises alm ost exclusively in children and young adults. Although not a com m on tum our by adult standards, its im portance lies in the age of population affected and its potential curability.
The role of radiotherapy in the m anagement of Ew ing' s sarcom a has changed completely since Jam es Ewing ® rst described a`diffu se endotheliom a of bone' . 3 Ewing' s sarcom a is essentially a system ic disease requiring system ic treatment. Before chemotherapy becam e availab le 20± 30 years ago, longterm survival rates w ere as poor as 10%. 4 O utcom e w as determined by m etastatic disease, w ith control at the prim ary site being of less im portance. Since the introduction of intensive com bination chemotherapy, survival has im proved considerably, with som e reports of long-term survival being as high as 50± 70%. 5,6 As chem otherapy has reduced death from m etastatic disease, local control (L C) has once again becom e m ore im portant.
In the m anagem ent of Ewing' s sarcom a, a number of unresolved issues persist in the integration of the treatment m odalities, including the roles of radiotherapy and surgery, and the scheduling of radiotherapy and chem otherapy. An additional important question which has attracted less attention is the optimal dose of radiotherapy required to achieve local control without in¯icting unacceptable norm al tissue dam age. Although Ewing him self observed response to radiotherapy, 3 there is com paratively little in the literature on this topic, and all m odern studies are com plicated by the fact that chem otherapy has a substantial effect on local disease. T his study has exam ined the effect of radiotherapy dose on LC , w ith the objective of contributing inform ation on dose± response in Ew ing' s sarcom a.

Patient details
All cases of Ewing' s sarcom a of bone referred to the Royal M arsden Hospital (RM H) during the period where N SD is expressed in rets, T is the overall treatment time in days and N is the num ber of fractions. It was assu m ed that for every course of radiotherapy including a com plete week, a weekend would also have been included, so that a 30-fraction course would have taken 42 days. The form ula was used to express treatm ents as equivalent total doses at 2 G y/fraction. H owever, the N SD m ethod incorporates considerable assum ptions. 8 T he indices in the equation were originally based on the study of acute-reacting`norm al' tissues, rather than tum ours. N evertheless, in at least one study, N SD was found to ® t clinical tum our data better than the linearquadratic m odel. 9 Although generally preferred, the m ore m odern linear-quadratic m odel does not, in its sim plest form , account for differences in overall treatment time and therefore cannot be app lied here to equate treatment schedules. 10

D ata analysis
Patients w ere categorized into those without m etastases at presentation (M 0) and those with m etastases (M 1). Patients with m etastases at presentation had a m uch shorter survival; the analysis has therefore focused on those patients who w ere M0 at presentation. 11 Although M 1 patients w ere analyzed, they have been treated separately. D ue to developm ents in diagnostic m ethods, especially C T scanning, it is likely that staging w as m ore accurate in the latter part of the study. Thus, it is possib le that som e patients classi® ed as M 0 early in the study would really have been M 1. For LC , a m inim um period of 3 m onths was left before recurrence was m onitored, to avoid confusion with failure to achieve control. T hus, patients dying within this period w ere excluded from calculations of LC. Patients were censored if am putation was perform ed for reasons other than local recurrence, or upon death, since they were no longer`available' to develop local recurrence.
In com paring the probability of events in different patient groups, the log-ran k test was used, with LC and survival probabilities described on a K aplan± M eier plot. W here several subsets of a param eter have a natural order, for exam ple radiotherapy dose in those patients receiving radiotherapy, a test of trend across groups w as applied, rather than a test of heterogeneity. Patients were adjusted for know n prognostic variables by strati® cation before com pari-1967± 1986 inclusive w ere identi® ed using the hospital database and cross-re ferenced against the H istopathology D epartm ent disease register. All histology was review ed at the RM H and only those cases con® rm ed as Ewing' s sarcom a w ere retained. F our small round cell tum ours of bone w ithout a m ore speci® c diagnosis and 10 Ewing' s sarcom as arising in soft tissue have been excluded from the analysis.
Inform ation on patients was obtained retrospectively as shown in T able 1. It proved im possible to collect details of tum our size or volum e. In the earlier years of the study, C T and m agnetic resonance im aging (M RI) were not available, so m easurem ents of the extent of the intram edullary com ponent and soft tissue extension could not be m ade. F or patients treated later, where scanning had been carried out, tumour size had not been recorded. Levels of lactate dehydrogenase (LD H) had not been m easured routinely. N o attem pt was m ade to detail the chem otherapy doses because it w as felt that this inform ation m ight be inaccurate w hen collected retrospectively. However, details of drugs used and intended doses, as well as the timing of chemotherapy in relation to radiotherapy and surgery, were collected.
For radiotherapy treatments, details of time, dose and fractionation were collected. All patients were treated with m egavoltage photons, except for four patients treated palliatively w ith low doses ( , 24 Gy, equivalent to 2 G y/fraction). The m argins of radiotherapy ® elds around the prim ary tum our could not be extracted retrospectively, partly because tum our extent could not be accurately assessed. A wide range of doses w as found, particularly in the ® rst decade of the series. However, the w ide variation in radiotherapy dose has allowed an attempt at evaluation of dose± response, which w ould have otherw ise been impossib le. relapse. Although m ost deaths occurred by 5 years, a few patients succum bed later; one patient died of disease in the 11th year after treatm ent. Of the 22 patients w ith m etastases at presentation, 21 are known to have died and the other patient was lost to follow-up w ith extensive disease. The LC rates for all patients were 71% (95% C I 60± 80%) at 2 years, 65% at 3 years and 56% (95% C I 41± 68%) at 5 years. T here was no difference in LC or survival according to sex or age at presentation, and no difference in either LC or survival between the ® rst and second decades of the study.
For M 0 patients the overall 5-year survival was 37% (95% CI 25± 50% ), and the 5-year disease-fre e survival rate 18% (95% C I 10± 28% T he survival fell to 22% at 8 years, and 16% at 10 years. It should be noted that the num ber of patients available`at risk' was sm all from 6 years on, but these late deaths have im plications for follow-up in future studies. T he L C rates for M 0 patients were 73% (95% CI 61± 83%) at 2 years, 70% at 3 years and 56% (95% C I 41± 70%) at 5 years (Fig. 2). T wo patients relapsed locally beyond 5 years, at 9 and 10.4 years, respectively, and both developed m etastases at the sam e time.
Eleven patients had symptoms attributable to tum our for over 2 years before presentation (the longest for 3.5 years). T here did not appear to be an increased risk of m etastases at presentation with longer duration of symptoms, and there was no difference in LC or survival.
Com paring duration of symptoms, there w as no difference in survival or LC . Eleven patients had symptoms attributable to tum our for over 2 years before presentation (the longest for 3Ã years). T here did not appear to be an increased risk of m etastases at presentation w ith longer duration of sym ptom s.
For all patients, the site of prim ary tum our had a highly signi® cant effect on overall survival (p 5 0.001, adjusted for extent of disease at presentation), but not on LC, categorizing prim ary site as: lim bs, axial skeleton or pelvis, in order of reducing survival. For M 0 patients, site had a signi® cant effect on survival (p 5 0.04). The relative risks of local failure for the axial skeleton, lim bs and pelvis were 1, 1.6 and 2.2, respectively. These differences were not statistically signi® cant, because of the sm all num ber of local failure events (26 in this group), although the m agnitude of the effect is consistent with a clinically im portant difference.
The adm inistration of chem otherapy appeared to have a potent effect on LC, reducing the relative risk of local relapse in M 0 patients to 0.53 (95% C I 0.22± 1.26%) (Fig. 3). T his was not statistically signi® cant (p 5 0.14), but very few patients did not receive chem otherapy, so the num bers in this comparison are sm all, and non-receipt of chem otherapy son. W here possib le, results have been adjusted for confounding variables, although it is unlikely that the effects of such factors have been com pletely rem oved. F or exam ple, the choice of treatment and dose, either radiotherapy or chem otherapy, m ay have have been in¯uenced by expected prognosis. D ue to the relatively small size of the cohort, full analysis via m ultivariate m ethods was not possib le. H azard ratios are referred to as relative risks throughout.

Results
N inety-six patients were available in the study cohort. Patient characteristics and the treatments they received are shown in T able 2, and the distribution of age in Fig. 1  m ay have been related to w idespread disease. In this series, there was no discernible difference in L C or survival w ith the num ber of chem otherapy agents adm inistered. High-dose chem otherapy with bone m arrow transplantation was not seen to improve LC or survival but this treatm ent w as reserved for patients w ith very extensive disease at presentation or following relapse. N o patients w ere successfully salvaged after relapse, although one patient survived 5 years.
Surgical resection in M 0 patients appeared to im prove LC , with the relative risk of local failure falling to 0.74 (adjusted for prim ary site) follow ing successful rem oval, but num bers were very sm all.
The distribution of radiotherapy dose is shown in Fig. 4, w ith total doses converted to be equivalent to 2 G y/fraction. O f the 17 patients in the > 60 Gy group, 10 received exactly 60 Gy in 30 daily fractions over 6 weeks; the highest dose delivered was 66 G y. Radiotherapy resulted in an im proved proba-  However, this effect on survival is likely to have been caused by om ission of radiotherapy from pa-tients with w idespread disease at presentation. T he same argum ent applies to the effect of radiotherapy on LC . Om ission of radiotherapy, or the delivery of a low dose, greatly reduced the probability of LC. LC rates with different radiotherapy doses for M0 patients are shown in F ig. 5. T he log-rank test for trend show s a statistically signi® cant effect for radiotherapy dose (p 5 0.04), shown in T able 3. A dose of , 40 Gy is asso ciated w ith a considerably re-duced chance of LC, and m ay not even be suf® cient for palliation in som e cases. All local failures in this group occurred within 1 year of treatment. A comparison of the three groups treated with > 40 G y dem onstrated no evidence of a dose-response (test for trend p 5 0.82), although very few local failure events w ere seen (only 19 in the M 0 group).

D iscussion
Progress in the m anagement of Ewing' s sarcom a of bone over the last 20 years has led to a dram atic im provem ent in prognosis, particularly for those patients who are m etastasis free at presentation. T his has refocused attention on the control of local disease. 12 Ewing originally described the tum our as being`highly susceptible to radium ' . 3 However, a signi® cant local recurrence rate rem ains. T his has led to the increased use of surgical resection, with radiotherapy reserved to follow incom plete surgery, or w hen surgical resection is im possible. T he radiotherapy question which has received m ost attention in recent years is scheduling with chem otherapy. H ow ever, the issues of radiotherapy dose, and the dose± response characteristics of this tum our rem ain poorly de® ned.
The purpose of this study was to investigate the effect of radiotherapy dose on local control. T here w ere 96 patients, which at ® rst sight appears a reasonable num ber. However, the high death rate and losses to follow -up substantially reduced the num ber of available local failure`events' , particularly when adjusting analysis for possible confounding factors.

Problems of retrospective studies
D rawbacks of retrospective studies include failure to collect data w hich are now known to be of prognostic value, such as tum our volum e, the dif® culty of identifying radiotherapy m argins around the tumour, and incom plete records of the reasons underlying m anagement decisions. T his can lead to analysis being confounded, w here the outcom e of interest has itself in¯uenced the choice of treatment.
The time period covered by the study saw considerable changes in the m ethods available for investigating and staging patients, w ith the introduction of CT and MRI. This did not cause any apparent problems in the analysis but was partly responsible for the increased proportion of M 1 patients seen in the second decade of the study (10% rising to 38%), and m ay have affected m anagement decisions. D ram atic changes also took place in chem otherapy schedules, surgical techniques, and planning and delivery of radiotherapy. All these factors have the effect of increasing heterogeneity in the study group, ham pering the interpretation of data. However, provided the danger of over-interpretation is avoided, the advantage of a study of this sort is the collection of inform ation on dose± response, without the use of a two dose-level random ized clinical trial.

Study results
T he distribution of age, sex, type and duration of symptoms, and the distribution of prim ary site in our series were typical (Fig. 1, T able 2). 2 Age had no effect on outcome. T his is consistent with other studies, although in som e, older age has been asso ciated w ith a w orse prognosis. 12,13 The site of the prim ary tum our proved of m ajor prognostic signi® cance, in accordance w ith other reported experience: limb sites carry a better prognosis than axial skeleton tum ours, w hich in turn are better than pelvic tumours. 12,14± 16 It m ay be that site is really a re¯ection of the size of tum our at diagnosis rather than being an independent prognostic variable. 17 In the Cooperative Ew ing' s Sarcoma Study (CESS) 81 trial, a Cox regression analysis identi® ed tumour volume ( , or > 100 cm 3 ) and histological response to initial chemotherapy as the m ajor determinants of prognosis. The prim ary site was not an independent prognostic factor, probably because of the link to tumour volum e. 17 The presence or absence of m etastases at the time of presentation is a m ajor prognostic factor. 12,15,17 T his emphasizes the im portance of initial staging, and the effect that a change in quality of staging investigations has on com parative results. The poor outlook of M 1 patients w as the reason for our focusing on m etastasis-fre e patients for the assessm ent of LC. C hem otherapy had a profound effect on LC (Fig. 3). H owever, no patients were salvaged after relapse, which is a m anifestation of the failure of chemotherapy to sterilize bulky disease.

Survival and local control
T he rates of both overall survival and LC in the study were disapp ointing com pared to current studies. It is reasonable to expect patients treated now to have long-term survival rates of around 50%, but such success has been achieved only com paratively recently. 5,11,17 O ur results are com parable with other studies reporting on patients treated from the 1960s to the early 1980s. 12,13,18,19 O nly the Intergroup Ew ing' s Sarcom a Study (IE SS-I) trial which recruited from 1973 to 1978 reported a substantially higher 5-year survival of 65% , but this was in patients with localized disease at presentation. 5 In IESS-II, recruiting from 1978 to 1982, patients with localized disease excluding pelvic prim aries had an overall 5-year survival of around 70%. 20 The CESS 81 and CE SS 86 trials have reported 3-year survival rates for patients with localized disease of 55% and 62% , also dem onstrating the im provem ent in survival which has becom e possible in the last decade. 17 LC rates have also been rising in m ore recent studies, reaching 3-year LC rates of around 70± 90% . 5,12,17 T he local failure rate in CESS 81 was around 50% and is thought to relate partly to poor radiotherapy planning. 6,17 Those studies which cover the early period of chemotherapy, through the 1960s and 1970s, generally report lower rates of LC , in the range of 40± 50% , and our results are consistent w ith these. 16,19 D uration of follow -up In our study, there was a sm all rate of attrition due to relapsing E wing' s sarcoma extending out to 11 years, w hich is the typical experience of studies with long follow-up. 5,12,18 It has occasionally been advocated that disease-fre e status at 5 years equates to cure, im plying that follow -up need not be continued beyond this, 13 but to assess true rates of cure certainly requires longer follow-up. T he C ESS studies have tended to report 3-year ® gures, because there are m any m ore`at risk' patients and m ost events occur within the ® rst 3 years. 17 T here is also a signi® cant incidence of second tum ours, partly related to an underlying predisposition in patients w ith Ewing' s sarcom a and partly to treatment, and there is an app reciable incidence of treatm entrelated com plications, som e of which are fatal. 13,21 It is therefore m andatory for follow -up to be long enough to record these events.

Radiotherapy dose± response
In the pre-chem otherapy era, it was noted that doses of , 40 G y resulted in frequent local failure, even though long-term survival was low. 4 M ore recent studies have failed to demonstrate a dose± response above 40 G y, although it is generally accepted that higher doses im prove LC . 6,17 O ur data are entirely consistent w ith these reports. In the CESS 81 study w here patients were random ized to receive either 46 Gy or 60 G y, no dose± response was found, with local failure just as frequent in the higher dose group. However, LC rates w ith radiotherapy were poor in this trial until centralized planning was established; this m ay have confounded any dose effect w hich m ight have been present. 6,17 Lack of dose± response has also been seen in other studies. 14,16 In the IE SS-I trial, it w as felt that this m ight have been due to the high incidence of death from m etastatic disease (alm ost 50% at 3 years) precluding clinical m anifestation of local recurrence. 11 Som e justi® cation for higher doses has com e from one study in which LC of bulky tumours ( > 100 cm 3 ) was im proved by doses of 55± 60 G y. 22 D oses above 60 G y, in com bination w ith chemotherapy, appear to offer no advantage in LC and have led to im paired functional outcom e from norm al tissue dam age. 23 T he consensus is that doses up to about 60 Gy are required for macroscopic dis-ease, although 45 Gy is considered to be adequate for m icroscopic disease. 6,17 It is possible that a dose± response does exist but that the search for it has been confounded by sm all num bers and technical problems. 13,14,17 In addition, a wide variation in intrinsic cellular sensitivity can lead to dif® culty in establishing a dose± response. Although in vitro data for Ewing' s sarcom a are limited, there is a m arked spread in sensitivity betw een tum ours. 24 Assim ilating the results from our study and from m any others, there is good evidence that doses below 40 G y are ineffective. In our study, in patients who received no radiotherapy or doses less than 40 G y, all local failures occurred within 1 year of treatment. This suggests that reasonably high doses are required even for palliative treatment, since patients with m etastatic Ewing' s sarcom a m ay survive for m any m onths.

C onclusions
In our study, radiotherapy im proved the probability of local control. O m ission of radiotherapy or a dose of less than 40 G y proved ineffective for LC, so that low doses m ay not necessarily be suf® cient for palliation. In the dose range 40± 66 G y, there was no evidence to suggest a dose± response.

A cknowledgem ents
W e would like to thank Miss Jane Regan for obtaining follow-up information and D r C yril F isher for reviewing the pathology. W e are also grateful to Professor Ann Barrett, D r Anna Casson i and D r M ark Gaze for stim ulating discussions. T he patients in this series w ere referred from many centres, and we are grateful to all those w ho have furnished updated follow-up inform ation. T he Institute of C ancer Research receives ® nancial support from the C ancer Research Cam paign.